Artificial intelligence trading agents face a critical credibility test. Researchers have developed BlindTrade, a framework that anonymizes stock tickers and company names to verify whether AI-powered trading systems genuinely understand market dynamics or simply memorize patterns from training data. The study tested four large language model agents on 2025 year-to-date performance, achieving a Sharpe ratio of 1.40 across multiple trials. The key finding reveals that AI trading signals remain valid even without ticker identifiers, suggesting legitimate market understanding rather than exploited memorization bias. However, performance varies significantly by market condition, excelling during volatile periods but underperforming in sustained bull markets. This research addresses fundamental concerns about deploying autonomous trading systems responsibly.
Post from MarketNews_en
Log in to interact with content.